"Is It Safe?" That’s the Wrong Question to Ask About AI

There's a famous exchange in C.S. Lewis's The Lion, the Witch and the Wardrobe where Lucy asks about Aslan: "Is he safe?" Mr. Beaver's reply is one of the most quoted lines in all of Lewis: "Who said anything about safe? But he is good."

That question - is it safe? - is the wrong question to ask about artificial intelligence. And understanding why might be the most important thing that faith leaders do this decade.

The Usefulness Paradox

Here's a principle worth naming: the more capable a tool is, the more dangerous it tends to be. The safer a tool is, the more limited its usefulness tends to be. Call it the Usefulness Paradox.

Consider the fixed blade knife. Over 1,500 people were murdered with one last year.  But that same tool was used to cut billions of restaurant steaks. Or the automobile: 1.2 million people died in car accidents globally last year, one person every two seconds. Yet cars also carried families to church, workers to jobs, and friends to one another across trillions of miles. No one is calling for the abolition of the kitchen knife or the car. We've learned to live with powerful tools because their benefits to human flourishing outweigh their risks, and we've built structures to manage the danger.

AI is no different; only the stakes are higher. Since 2023, AI has been linked to at least a dozen suicides and to thousands of lives saved through medical breakthroughs in diagnostic imaging and drug discovery. As with every tool humanity has ever wielded, AI simply amplifies what is already in the human heart. That's what makes it a paradox, not a problem to be solved with a disclaimer.

The Real Question: Trusted

If "safe" is the wrong question, what's the right one? “Is it trustworthy?”

No powerful tool can be made completely safe, but it can be deliberately oriented. Seat belts and airbags don't eliminate the danger of cars. They save roughly 19,000 lives per year in the U.S. by managing it. The question isn't whether AI can cause harm. It's whether the values embedded in a given AI system are genuinely oriented toward your good.

This matters because not all AI is created equal. OpenAI's mission statement used to read: "To build general-purpose AI that safely benefits humanity." In 2024, as the company restructured into a for-profit entity, they quietly removed the word "safely." Words matter. So do the actions behind them.

What Gloo Is Actually Doing

At Gloo, we believe the faith & flourishing ecosystem deserves better than vague safety language. That's why we built the Flourishing AI Initiative (FAI), the first evaluation framework of its kind that measures AI not by capability, but by whether its responses actively help people flourish.

FAI assesses AI models across seven research-backed dimensions drawn from Harvard's Human Flourishing Program and the Global Flourishing Study: character, happiness, relationships, meaning, health, finances, and faith. In our December 2025 evaluation, none of the 28 models tested reached our 90-point flourishing threshold. When we introduced FAI-Christian, which evaluates how well AI models align with a Christian worldview, scores dropped even further, ranging from the low 50s to the high 70s across all frontier models tested. 

The gap is real. And it's why this work matters.

Trustworthy, Not Perfect

The Usefulness Paradox doesn't paralyze us. It clarifies our calling. We're not trying to build a rubber knife. We're building toward something more like the autonomous vehicle: AI that is architecturally oriented toward protecting human dignity, that treats flourishing as a design requirement rather than a PR talking point.

Mr. Beaver was right. The question was never whether Aslan was safe. The question was whether he was good.

That's the question we're asking and answering at Gloo.

Want to explore the data? Visit gloo.com/flourishing-hub/research to see how today's leading AI models measure up against the dimensions of human flourishing.